University of Houston

Detection and morphology extraction of 3D Tubular Structures (TS) in biomedical images is important for a variety of applications (e.g., analysis of biological structures, diagnosis, monitoring of diseases, and surgical planning). In the domain of neuroscience, robust and accurate methods for the automatic extraction of neuronal morphology from 3D image stacks are necessary in order to quantify phenotypic changes of neurons. Imaging data from neuronal networks pose numerous challenges including: (i) low contrast; (ii) possibly high levels of noise, (iii) large difference in the size of the structures that need to be identified, and (iv) crossing and possibly touching dendrites.

To address this critical need, we propose the development of new methods for the segmentation of tubular structures and centerline extraction. The specific objectives include: (i) develop a framework for the generation of computational phantoms that can be used to evaluate segmentation algorithms of regular tubular structures, (ii) develop and evaluate a segmentation algorithm able to segment regular tubular structures; (iii) develop and evaluate a segmentation algorithm able to segment irregular tubular structures; (iv) develop a centerline generation algorithm robust to the challenges of outlined above. These novel methods will be incorporated at the ORION software to facilitate rapid and objective processing of large-scale multispectral fluorescence images of complex neuronal networks.